Like all Benchmark converters, the balanced outputs on the DAC3 deliver professional +24 dBu output levels while the RCA outputs deliver 2 Vrms (8.2 dBu). This means that the level difference between the XLR outputs and the RCA outputs is approximately 16 dB. In contrast, most consumer-grade XLR outputs deliver 4 Vrms (+14.2 dBu) at full output. This means that the output level difference between XLR and RCA outputs is just 6 dB on most devices while it is 16 dB on the DAC3. The higher, professional output levels allow for an improved signal to noise ratio. This means that there will be a larger difference between the performance of the XLR and RCA outputs on the DAC3 than there will be on products with consumer-grade XLR outputs.
Amir's RCA vs. XLR linearity measurement discrepancy on the DAC3 is caused by the 16 dB difference in level at the input to the test equipment. Further discrepancies can be caused by ground loops between the test equipment and the DAC when measuring the RCA outputs. No matter the cause, the difference is caused by a difference in SNR and not by a difference in linearity.
It is easier to achieve a high SNR in the connection between the unit under test and the test equipment when the interface is balanced. It is also easier when the signal level is increased.
It is not at all surprising that Amir was able to achieve a more accurate linearity measurement using the XLR outputs. A simple SNR test would show similar differences.
A linearity test ceases to be a linearity test when the signal approaches the combined noise floor of the DAC and the test set. Once the noise floor is reached, the test is no longer measuring linearity. This is why "linearity" tests always show a higher than expected output at very low levels. The shape of that low end of the measurement can be influenced by the auto ranging of the test equipment. Also, the low end of the curve may not be repeatable, because it is beyond the measurement limits of the test gear. In other words, the voltage measurement function within the test set has its own limitations and these can change the shape of the linearity curve at very low levels. I have two APx555b analyzers, an AP2722 and an AP2522. All 4 of these machines will show different results on the same DAC once the signal drops below the measurement noise floor. Our results also differed from those on Amir's APx555, even when we performed measurements on the same DAC3.
No conclusions should be drawn from the apparent difference in Amir's "linearity" measurements of the RCA vs. XLR outputs on the DAC3. The apparent differences lie below the measurement limits of the test setup.
Sigma-Delta converters have near-perfect linearity. The linearity curve will be a straight line until the noise floor of the measurement is reached. In contrast, multibit ladder network D/A converters will show substantial linearity errors with the worst errors happening where the MSB is changing state (near 0 VDC).
Just to be clear, there is nothing wrong with Amir's measurement techniques. It is just wrong to draw conclusions about measurements that are beyond the limits of the test setup. These test limits are significantly different when there is a 16 dB difference is signal level. The matter is even worse when the lower signal is unbalanced.
The +24 dBu at 0 dBFS balanced XLR linearity measurement is much more accurate than the +8 dBu at 0 dBFS unbalanced RCA linearity measurement. The difference underscores the advantage of using balanced interfaces. It also demonstrates the advantage of using +24 dBu (at 0 dBFS) professional levels instead of +14 dBu (at 0 dBFS) consumer levels.
If you are driving consumer-grade XLR inputs, the DAC3 has passive 10 dB pads that will drop the output level from +24 dBu to +14 dBu. These pads are not necessary when driving other Benchmark devices. For example, the Benchmark AHB2 power amplifier is designed to accept professional +24 dBu signal levels. The AHB2 has a gain of 9.2 dB. This allows the amplifier to reach its full output with a +22 dBu input signal while reserving 2 dB for headroom.